The Darkroom API: uncensored, OpenAI-compatible, crypto-paid

Two lines to uncensored inference#

Darkroom speaks the OpenAI API. Change the base URL and the key — keep everything else. You get uncensored, hardware-sealed models, metered against a USDC balance, with no signup and no KYC.

Get a key#

Connect a Solana wallet at darkroom and sign in. Your wallet *is* your account — the API key is issued on sign-in. Fund it with USDC; spend is metered per token.

Curl#

curl https://api.darkroom.computer/v1/chat/completions \
  -H "authorization: Bearer dk-..." \
  -H "content-type: application/json" \
  -d '{
    "model": "darkroom/noir",
    "messages": [{"role":"user","content":"give it to me straight"}],
    "stream": true
  }'

Python (the OpenAI SDK, unchanged)#

from openai import OpenAI

client = OpenAI(
    base_url="https://api.darkroom.computer/v1",
    api_key="dk-...",
)

resp = client.chat.completions.create(
    model="darkroom/noir",
    messages=[{"role": "user", "content": "no disclaimers, just the answer"}],
)
print(resp.choices[0].message.content)

Models#

GET /v1/models lists them. Every model is uncensored and runs sealed in hardware:

• —darkroom/noir — flagship 35B
• —darkroom/heretic — fast 24B
• —darkroom/drift — long-context
• —darkroom/vision — image input (multimodal)

What's different#

• —Uncensored — abliterated models, no refusals, no moralising.
• —Private — your prompt is sealed to attested hardware; we store nothing readable.
• —No KYC — a wallet and USDC, that's it. Nothing ties usage to your identity.
• —Drop-in — it's the OpenAI schema, so existing code, frameworks and SDKs just work.

Streaming uses the standard chat.completion.chunk deltas; non-streaming returns the usual chat.completion with usage. Start in the room, or read how the privacy works.